Gate to Computer Vision and Pattern Recognition (gtCVPR) – Volume 2 Number 1 – 2020

Abstract

In the post segmentation scenario, when objects in the scene have been extracted, the focus shifts to object identification. This can be achieved through shape or texture. Finding the object boundary has been a reliable means of shape description. Among the mathematical approximation techniques for shape analysis, Fourier descriptors have proven to approximate closed boundaries of objects quite well, albeit with some limitations. A statistical thresholding technique to restrict the number of descriptors for a reasonably good approximation of the target shape is explored and tested on some medical images. Encouraging results were obtained particularly when segmentation in the pre processing stage was effectively carried out.